End of training

README.md

@@ -62,14 +62,14 @@ lora_alpha: 16
 lora_dropout: 0.05
 lora_fan_in_fan_out: null
 lora_model_dir: null
-lora_r: 8
+lora_r: 32
 lora_target_linear: true
 lr_scheduler: cosine
-max_steps: 100
+max_steps: 500
 micro_batch_size: 1
 mlflow_experiment_name: /tmp/dcd10050e81faec5_train_data.json
 model_type: AutoModelForCausalLM
-num_epochs: 1
+num_epochs: 4
 optimizer: adamw_bnb_8bit
 output_dir: miner_id_24
 pad_to_sequence_len: true
@@ -77,7 +77,7 @@ resume_from_checkpoint: null
 s2_attention: null
 sample_packing: false
 saves_per_epoch: 1
-sequence_len: 512
+sequence_len: 1024
 strict: false
 tf32: false
 tokenizer_type: AutoTokenizer
@@ -102,7 +102,7 @@ xformers_attention: null
 
 This model is a fine-tuned version of [unsloth/tinyllama](https://huggingface.co/unsloth/tinyllama) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 2.1172
+- Loss: 2.0288
 
 ## Model description
 
@@ -130,13 +130,17 @@ The following hyperparameters were used during training:
 - optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_steps: 10
-- training_steps: 100
+- training_steps: 500
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 2.1574        | 0.0249 | 100  | 2.1172          |
+| 2.9282        | 0.0002 | 1    | 3.6463          |
+| 1.7301        | 0.0311 | 125  | 2.0661          |
+| 2.1628        | 0.0622 | 250  | 2.0405          |
+| 2.3675        | 0.0933 | 375  | 2.0310          |
+| 1.958         | 0.1244 | 500  | 2.0288          |
 
 
 ### Framework versions

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b4b23a79a04d4a7a5940cdfe0a6029ffe3af055d536b798ef0c3dc1f81a0385
-size 25342042
+oid sha256:753d8cee6d7ba4c42f760b97f929dd77e9e3298fee2ae3d2071a443ee83652ae
+size 101036698